Robot state prediction method and system based on digital twinning

Abstract

The application discloses a robot state prediction method and system based on digital twinning. By introducing the decline prediction trend and unified risk assessment, the problems of prediction inaccuracy and decision fragmentation are solved. On the one hand, the continuous change curves of stiffness, viscous friction and Coulomb friction coefficient in the preset time window are extrapolated by using historical decline parameters, and are injected into the digital twin to perform advanced deduction, so that the predicted trajectory can reflect the continuous evolution of the decline, thereby eliminating the parameter freezing deviation and improving the long-period prediction accuracy. On the other hand, the trajectory deviation curve and the collision risk evolution curve are comprehensively evaluated, and a threshold is set to realize the classification decision: when the trajectory deviation exceeds the limit but the collision risk does not exceed the limit, compensation control is performed; when the collision risk exceeds the limit, emergency collision avoidance is preferentially executed, forming a mutually exclusive safety intervention mechanism, avoiding the instruction conflict or intervention delay caused by independent processing of internal and external risks, and guaranteeing the operation accuracy and safety.

Description

Technical Field

[0001] This application relates to the field of robotics technology, and in particular to a method and system for predicting robot state based on digital twins. Background Technology

[0002] With the widespread application of industrial robots in high-end manufacturing, the problem of mechanical degradation during long-term operation is becoming increasingly prominent. Electromechanical losses such as wear of joint bearings, increased gear backlash, and deterioration of lubricating grease lead to a growing deviation between the robot's actual operating trajectory and its nominal commands, severely impacting processing accuracy and product quality. Simultaneously, robots coexist with static obstacles and dynamic personnel and equipment in complex workspaces, making collision risk a persistent key factor restricting their safe operation. Therefore, how to perceive the robot's mechanical degradation status in real time and proactively predict its future operating trajectory and collision risks is a core issue in achieving predictive maintenance and proactive safety control of robots.

[0003] In existing technologies, parameter identification methods based on robot dynamics models are commonly used for state monitoring. A typical approach involves collecting real-time operational data such as joint torques and vibrations to assess the current level of mechanical wear and update the corresponding parameters in the dynamics model accordingly, ensuring the model's state matches that of the physical robot. Building upon this, some solutions input the anticipated motion commands into the updated dynamics model for simulation, obtaining predicted trajectories for collision detection.

[0004] However, existing technologies still have the following shortcomings: First, when extrapolating future trajectories, model parameters are usually frozen at the current identified values, without considering that mechanical decay continues to evolve within the prediction time window, resulting in the accumulation of deviations between the predicted and actual trajectories over time and insufficient prediction accuracy; Second, existing solutions mostly treat trajectory deviation and collision risk as two independent modules, lacking a mechanism for comprehensive assessment and unified decision-making on internal decay risk and external collision risk, making it difficult to generate coordinated safety intervention strategies in different risk intersection scenarios. Summary of the Invention

[0005] The purpose of this application is to provide a robot state prediction method and system based on digital twins, which solves the problems of inaccurate trajectory prediction caused by ignoring the evolution of decline trends and the separation of internal and external risk decision-making in the prior art.

[0006] The first aspect of this application provides a robot state prediction method based on digital twins, the method comprising: Acquire real-time operational data of the robot and the physical environmental elements of its workspace; The time-varying mechanical decay of the robot is evaluated based on the real-time operating data, and the decay parameters in the digital twin are updated to obtain a digital twin that is synchronized in real time with the current electromechanical loss state. The digital twin is a digital model pre-built in the edge controller that corresponds to the mechanical dynamics properties of the robot. The robot's expected motion commands within a preset time window are obtained. The decay prediction trend within the preset time window and the expected motion commands are input into the digital twin for advanced extrapolation, and the predicted running trajectory and trajectory deviation curve are calculated. The decay prediction trend is obtained by extrapolation from the digital twin based on historical decay parameters. Based on the physical environment elements, a spatiotemporal constraint boundary is constructed in the virtual workspace, and the predicted running trajectory is mapped to the virtual workspace for collision risk detection, generating a collision risk evolution curve. A comprehensive state assessment is performed based on the trajectory deviation curve and the collision risk evolution curve. When the operational risk is determined to exceed the preset safety threshold, the compensation control quantity and collision avoidance bias quantity are calculated and injected into the motion control loop to achieve safe intervention for the robot.

[0007] A second aspect of this application provides a robot state prediction system based on digital twins, the system comprising: The acquisition module is used to acquire the robot's real-time operating data and the physical environmental elements of its workspace; The decay parameter update module is used to evaluate the time-varying mechanical decay degree of the robot based on the real-time operating data and update the decay parameters in the digital twin to obtain a digital twin that is synchronized with the current electromechanical loss state in real time. The digital twin is a digital model pre-built in the edge controller that corresponds to the robot's mechanical dynamics attributes. The calculation module is used to obtain the robot's expected motion command within a preset time window, input the decay prediction trend within the preset time window and the expected motion command into the digital twin for advanced extrapolation, and calculate the predicted running trajectory and trajectory deviation curve; wherein, the decay prediction trend is obtained by extrapolation from the digital twin based on historical decay parameters; The collision risk detection module is used to construct a spatiotemporal constraint boundary in the virtual workspace based on the physical environment elements, map the predicted running trajectory to the virtual workspace to perform collision risk detection, and generate a collision risk evolution curve. The risk control module is used to perform a comprehensive state assessment based on the trajectory deviation curve and the collision risk evolution curve. When it is determined that the operating risk exceeds the preset safety threshold, the module calculates the compensation control quantity and the collision avoidance bias quantity and injects them into the motion control loop to achieve safe intervention for the robot.

[0008] Compared with the prior art, the beneficial effects of the present invention are at least as follows: The technical solution provided in this application effectively solves the problem of inaccurate prediction and disconnect between safety intervention in existing technologies by introducing a unified decision-making mechanism for decay prediction trends and internal and external risks. On the one hand, the edge controller uses the accumulated decay parameter updates within historical operating cycles to extrapolate the continuous change curves of stiffness attenuation factor, viscous friction correction coefficient, and Coulomb friction correction coefficient within a preset time window through a time series prediction algorithm. These curves are then injected as time-varying dynamic parameters into the dynamic equations of the digital twin, enabling the predicted operating trajectory to reflect the continuous evolution of mechanical decay on the prediction time axis. This avoids the accumulation of prediction deviations caused by parameter freezing and significantly improves the accuracy of long-cycle trajectory prediction. On the other hand, by incorporating the trajectory deviation curve and the collision risk evolution curve into a unified comprehensive state assessment framework, and setting trajectory deviation thresholds and collision risk thresholds, the system classifies and determines two risk states: "accuracy decay" and "collision crisis." When the trajectory deviation exceeds the limit but the collision risk does not, compensation control is triggered; when the collision risk exceeds the limit, emergency collision avoidance is triggered first. This hierarchical and mutually exclusive decision-making mechanism avoids the problems of conflicting control commands or delayed intervention caused by decision-making based on internal and external risks, thus ensuring the robot's operational accuracy and inherent safety in complex human-machine collaborative environments. Attached Figure Description

[0009] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0010] Figure 1 This is a flowchart of a robot state prediction method based on digital twin in an embodiment of this application; Figure 2 This is a structural diagram of a robot state prediction system based on digital twins, as described in an embodiment of this application. Detailed Implementation

[0011] For ease of understanding, the specific process of the embodiments of this application is described below. Please refer to [link / reference]. Figure 1 This application provides a robot state prediction method based on digital twins, comprising: Exemplary embodiments of this application will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of this application are shown in the drawings, it should be understood that this application may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of this application and to fully convey the scope of this application to those skilled in the art.

[0012] It should be noted that, unless otherwise stated, the technical or scientific terms used in this application shall have the ordinary meaning as understood by one of ordinary skill in the art to which this application pertains.

[0013] The methods described in the embodiments of this application will be explained in detail below.

[0014] This embodiment provides a robot state prediction method based on digital twins. This method can be applied to high-precision trajectory tracking tasks involving human-robot collaboration or complex environments in scenarios such as large factories, hospitals, and warehousing and logistics centers. In this embodiment, the core computing tasks can be flexibly deployed on the edge controller mounted on the robot body, or completed collaboratively by the edge controller and the cloud server through a wireless communication network, depending on the actual needs of the scenario. For example, the construction of the digital twin, the updating of decay parameters, and the advance simulation can be deployed on the edge side to achieve low-latency real-time response, while the offline training of historical decay trends and the fusion of global obstacle information can be deployed on the cloud side to reduce the edge computing load.

[0015] Step 110: Obtain the robot's real-time operating data and the physical environmental elements of its workspace; This step serves as the data source for the entire state prediction and control method. Through multi-source sensor collaborative sensing, it synchronously collects the internal state data of the robot body and the dynamic and / or static environmental elements of the external workspace, providing a basic data stream for subsequent digital twin updates and spatial collision prediction.

[0016] As a preferred implementation method, the specific acquisition and composition of real-time operational data and physical environment elements are as follows: Real-time operational data is used to characterize the kinematic, electrical, and thermodynamic states of the robot joints in real time. This includes joint angular displacement, joint angular velocity, joint contact temperature, and servo motor q-axis current collected by robot joint sensors and servo drivers, as detailed below: Joint angular displacement and angular velocity: High-frequency acquisition is performed using encoders integrated at the ends of the motor shafts of each joint of the robot to obtain the current angular displacement and angular velocity of each joint. This kinematic data is subsequently used to input the robot's ideal dynamic equations to calculate the theoretical torque under healthy conditions.

[0017] Joint contact temperature: acquired in real time via a contact temperature sensor attached to the robot reducer housing or motor surface. Since the viscosity of the lubricating grease inside the reducer is extremely sensitive to temperature, this temperature data is subsequently used to correct torque deviations and eliminate frictional fluctuations caused by changes in grease viscosity due to temperature rise.

[0018] Servo motor q-axis current: The q-axis current of the motor is directly obtained by reading the current feedback register inside the servo driver in real time via the industrial bus. Since the electromagnetic torque of the motor is directly proportional to the q-axis current, this current data not only serves as a direct basis for calculating the actual torque of the robot, but its high-frequency pulsation signal is also used to analyze the nonlinear vibration characteristics caused by mechanical gear wear.

[0019] Physical environment elements are used to reconstruct the boundary constraints of the physical world in virtual space. These elements include the spatial coordinates of static obstacles and the motion trajectories of dynamic obstacles, which are collected in real time by external 3D sensing devices deployed in the workspace. Specifically, these elements include: Static obstacle spatial coordinates: The workspace is scanned by a 3D stereo vision camera or LiDAR to obtain 3D point cloud data of fixed equipment in the space, and then converted into spatial geometric coordinates of static obstacles in the virtual workspace.

[0020] The motion trajectory of dynamic obstacles: Using external 3D sensing devices such as depth cameras or multi-line LiDAR deployed above the workspace, dynamic interference objects such as collaborating personnel and mobile AGVs entering the robot's work area are identified and continuously tracked in real time, generating their spatial motion displacement curves. This trajectory is subsequently used to construct spatiotemporal constraint boundaries that dynamically shrink or expand over time in virtual space.

[0021] Step 120: Evaluate the time-varying mechanical decay of the robot based on real-time operating data, update the decay parameters in the digital twin to obtain a digital twin that is synchronized in real time with the current electromechanical loss state. The digital twin is a digital model pre-built in the edge controller that corresponds to the robot's mechanical dynamics attributes.

[0022] In this embodiment, the digital twin is the core of mechanistic computation deployed in the real-time operating system of the robot's edge controller, rather than a graphical model used only for 3D visualization. Its function is to reproduce the dynamic behavior of the physical robot in virtual space with high fidelity, and to predict or calculate the robot's joint torques in advance through parallel computing, providing a computational basis at the physical mechanism level for subsequent control compensation, decay prediction, and collision avoidance.

[0023] Digital twins are constructed based on the rigid-flexible coupling dynamics equations of robots, and their standard mathematical structure is as follows:

[0024] in, Here is the mass inertia matrix. The matrix represents the centrifugal force and the Coriolis force. It is the gravity vector; For time-varying friction force models that include temperature effects, This model represents the joint flexibility deformation and stiffness decay. It uses the factory-calibrated dynamic parameters of the digital twin-loaded robot under factory reference conditions. As the physical robot gradually wears down during service, this embodiment identifies and corrects time-varying decay parameters (such as the friction coefficient) in the above equations online. The stiffness coefficient k enables the digital twin to synchronously evolve from its ideal factory state to its current real decay state.

[0025] Centrifugal force and Coriolis force matrix Characterizing the dynamic effects generated by the coupling of angular velocities at each joint during robot motion, its elements are derived from the mass-inertia matrix. The elements are obtained by combining the derivatives of the first kind of Christofel notation.

[0026] Time-varying friction model The mathematical expression used to describe the physical effect of contact temperature on grease viscosity and the effect of gear surface wear on frictional resistance is as follows:

[0027] in, It is a symbolic function.

[0028] Coulomb friction term This evolution, occurring with gear surface wear, can be expressed as:

[0029] The nominal coulomb friction force is specified by the manufacturer. This is the Coulomb friction correction factor.

[0030] Viscous friction term Due to the combined effects of temperature and aging deterioration of the lubricating grease, it can be expressed as:

[0031] in, The nominal viscous friction force is specified by the manufacturer. The correction factor is derived from the viscous friction coefficient. The temperature viscosity index is the temperature viscosity index of the grease. For real-time joint contact temperature, The reference temperature. The above temperature-viscosity correction term. The step is used to mitigate the influence of temperature on torque deviation: as joint temperature increases, this exponential term decreases, and viscous friction... The corresponding decrease means that the frictional fluctuations caused by temperature can be eliminated from the torque deviation through reverse calculation, thus obtaining the true mechanical wear amount.

[0032] Stiffness decay model The mathematical expression for characterizing the elastic deformation of a joint under load and its stiffness degradation after long-term service is as follows:

[0033] in, This is the elastic deformation angle between the motor end and the reducer output end caused by the force applied. (Time-varying stiffness coefficient) The degradation occurs with the fatigue wear of the flexible bearing in the reducer, expressed as: .in, This refers to the manufacturer's nominal stiffness coefficient. This is the stiffness attenuation factor, and its value range is... In the initial ideal state .

[0034] Step 120 uses the four data points—joint angular displacement, angular velocity, temperature, and q-axis current—obtained in step 110 to assess the robot's current physical degradation state. These parameters are physically highly coupled. In this embodiment, through physical decoupling and fluctuation analysis, the comprehensive degradation state is decomposed and updated to the corresponding mechanistic parameters in the digital twin.

[0035] In a specific implementation, step 120 can be achieved by the following steps: The theoretical torque is calculated based on the joint angular displacement and joint angular velocity, and the actual torque is calculated based on the q-axis current of the servo motor. The torque deviation between the two is then obtained. By combining the joint contact temperature, the interference caused by temperature fluctuations in the torque deviation is eliminated, and the stiffness degradation and viscous friction increment after eliminating the influence of temperature are obtained. High-frequency fluctuation characteristics in the q-axis current of the servo motor are extracted to determine the Coulomb friction increment caused by gear wear. Based on the stiffness degradation, viscous friction increment, and Coulomb friction increment, the corresponding model parameter update amounts are calculated and updated in the digital twin.

[0036] Specifically, the edge controller receives real-time sampled joint angular displacement and angular velocity, which are then substituted into the dynamic equations of the digital twin. At this point, the friction term in the equations... Use the factory-nominated parameters (i.e.) , Stiffness term Use the factory-specified stiffness coefficient (i.e.) The theoretical torque under ideal conditions is calculated. .

[0037] At the same time, the edge controller uses the currently collected servo motor q-axis current... Combined with torque constant And the reduction ratio, To determine the transmission efficiency, the actual output torque of the robot is calculated. :

[0038] The difference between the two torques is the torque deviation. :

[0039] This torque deviation, on a macroscopic level, includes the torque increment caused by all mechanical degradation factors such as increased friction and decreased stiffness.

[0040] Combined with joint contact temperature Eliminate torque deviation This method addresses the interference caused by temperature fluctuations, thereby accurately determining the stiffness degradation and viscous friction increments after eliminating the influence of temperature. The specific physical and computational logic is as follows: The viscosity of the grease inside the reducer decreases exponentially with increasing temperature, and the actual torque decreases due to temperature rise. This temperature-induced torque change is not a true physical wear phenomenon. This is based on the aforementioned temperature-viscosity model. The contribution of temperature to viscous friction is a relative exponential factor. The edge controller first utilizes the real-time measured joint contact temperature. Preset reference temperature and temperature viscosity index The current viscous friction temperature correction factor is calculated. Then, this correction factor is used to recalculate the nominal viscous friction force, and the torque deviation is considered. The reduction in viscous friction due to temperature rise is subtracted, thus eliminating the spurious torque fluctuations caused by temperature. After removing the influence of temperature, the remaining torque deviation is purely caused by physical wear.

[0041] For the viscous friction increment: Extract the residual torque deviation during the robot's stable, uniform speed operation phase (when acceleration is 0 and stiffness and elastic torque remain constant). This deviation is proportional to the joint angular velocity, and the current viscous friction increment (corresponding parameter) is determined accordingly. ).

[0042] For stiffness degradation: Extract the residual torque deviation under alternating load conditions such as robot acceleration, deceleration, or reversal. Since the joint experiences slight torsional deformation under stress during these phases, analyze the proportional relationship between this deformation and the torque deviation to determine the current stiffness degradation of the joint (corresponding parameter). ).

[0043] In summary, the edge controller combines the currently measured joint contact temperature Introducing a temperature-viscosity correction formula, from the torque deviation The change in viscous friction caused by temperature fluctuations during the peeling process yields the viscous friction increment purely due to physical wear, which characterizes the permanent chemical degradation or physical wear of bearings and greases. Based on this, under transitional conditions such as robot start-up, stopping, or high-load reversal, the torque deviation after the effect of peeling temperature is analyzed, the minute elastic deformation of the joint under load torque is calculated, and the stiffness degradation of the joint is determined, characterizing the decline in the flexible bearing or torsional stiffness of the RV reducer.

[0044] Extracting the q-axis current of the servo motor The high-frequency fluctuation characteristics are used to determine the Coulomb friction increment caused by gear wear. The specific physical and computational logic is as follows: When pitting, peeling, or increased meshing clearance occur on the surface of the reducer gears, periodic impact vibrations will be generated during gear meshing. This impact will affect the q-axis current of the motor. This manifests as high-frequency ripple. In this embodiment, this high-frequency fluctuation is quantified by extracting the standard deviation of the q-axis current in the time domain, or by extracting the vibration energy at the characteristic frequency of gear meshing using a fast Fourier transform. Since Coulomb friction mainly depends on the surface roughness of the gear and the degree of meshing interference, the increase in Coulomb friction caused by gear surface wear can be determined by the magnitude of the aforementioned high-frequency current fluctuation.

[0045] After obtaining the stiffness degradation, viscous friction increment, and Coulomb friction increment, the edge controller transforms these deviation physical quantities into nonlinear proportional coefficients in the dynamic equations, i.e., the corresponding model parameter updates (such as stiffness correction coefficients). Viscous friction correction coefficient Coulomb friction correction coefficient ).

[0046] These updates are written into the running digital twin equations, enabling the corresponding stiffness and friction parameters to iterate online. At this point, the digital twin within the edge controller achieves virtual-real synchronization, accurately reflecting the current true decay state of the physical robot.

[0047] Step 130: Obtain the robot's expected motion commands within a preset time window, input the decay prediction trend and expected motion commands within the preset time window into the digital twin for advanced extrapolation, and calculate the predicted running trajectory and trajectory deviation curve; wherein, the decay prediction trend is extrapolated by the digital twin based on historical decay parameters.

[0048] To accurately simulate the robot's actual physical state during future operation, this embodiment does not directly substitute the current decay state as a constant value, but instead considers the continuous evolution of decay parameters within the prediction time window. To obtain the robot's physical aging trend within the future preset time window, the edge controller extrapolates and generates a decay prediction trend through the following steps: Obtain the cumulative amount of decay parameter updates for the digital twin over its historical operating cycle; The historical sequence of the update of the recession parameters is fitted using a time series prediction algorithm, and the parameter change curves that evolve over time within a preset time window are extrapolated as the recession prediction trend.

[0049] Specifically, the edge controller retrieves the stiffness degradation amount stored in the historical deviation database. Increment of viscous friction and Coulomb friction increment The historical time series is obtained. Time series prediction algorithms (such as the first-order grey prediction model GM(1,1), quadratic polynomial fitting, or Holt-Winters double exponential smoothing algorithm) are used to fit the above historical series and extrapolate it to the future time axis to obtain the results within a preset time window. , , The curves change continuously over time. This set of curves represents the degradation prediction trend, describing the physical trend of the robot's joint stiffness continuing to weaken and friction continuing to increase during future tasks.

[0050] After obtaining the predicted decay trend, the edge controller performs advanced extrapolation in virtual space. In a specific implementation, the predicted decay trend and expected motion commands within a preset time window are input into the digital twin for advanced extrapolation to calculate the predicted trajectory and trajectory deviation curve. This can be achieved through the following steps: The decay prediction trend is substituted into the dynamic equation of the digital twin as a time-varying decay parameter. The expected motion command is used as input, and simulation is performed within a preset time window to obtain the predicted joint motion sequence, which is used as the predicted running trajectory. The deviation between the predicted trajectory and the expected motion command is calculated, and a trajectory deviation curve is generated as a function of time.

[0051] Specifically, the above will change over future time. , , The curve serves as a time-varying decay parameter, which is then injected into the dynamic equations of the digital twin. At this point, the digital twin becomes a nonlinear time-varying system whose parameters change continuously with the simulation time.

[0052] Using the expected motion commands (nominal angle, velocity, and acceleration) within a preset time window as input, a digital twin with injected time-varying parameters is driven to perform forward dynamics simulation. Utilizing the high-speed computing power of the edge controller, this simulation can complete the entire preset time window simulation in a very short time, outputting the simulated response angles of the virtual joints at various future moments, denoted as the predicted trajectory. The predicted trajectory will be deduced. With the expected motion command input By subtracting the values ​​and taking the absolute value, the deviation curve of the trajectory distribution on the future time axis can be calculated. :

[0053] This curve quantifies the time period, location, and magnitude of trajectory tracking deviations caused by electromechanical degradation during future operation of the robot.

[0054] Step 140: Based on physical environment elements, construct spatiotemporal constraint boundaries in the virtual workspace, map the predicted running trajectory to the virtual workspace for collision risk detection, and generate a collision risk evolution curve.

[0055] During robot operation, in addition to trajectory deviations caused by robot body decay, collaborating personnel or mobile devices within the workspace are also in dynamic motion. The purpose of this step is to place the predicted trajectory obtained in step 130 into a virtual scene containing both static environment and dynamic obstacles to perform advanced collision risk detection.

[0056] First, the edge controller overlays static environmental elements and dynamic obstacle information onto the timeline within the virtual workspace to construct a spatiotemporal constraint boundary. In a specific implementation, this construction process involves the following steps: Convert the spatial coordinates of static obstacles into static spatial constraint boundaries in the virtual workspace; Based on the movement trajectory of dynamic obstacles, predict the dynamic occupied area within a preset time window in the future, and construct a dynamic spatial constraint boundary that changes dynamically over time. By fusing static and dynamic spatial constraint boundaries on the time axis, a spatiotemporal constraint boundary is generated.

[0057] Transformation of static spatial constraint boundaries: The edge controller retrieves the geometric coordinates of static obstacles (such as fixed machine tools, conveyor belts, fences, etc.) obtained in step 110, and performs 3D wrapping in the virtual workspace using oriented bounding boxes or axis-aligned bounding boxes to generate static spatial constraint boundaries that remain constant on the time axis. .

[0058] Prediction and construction of dynamic space constraint boundaries: For the current motion trajectory of the collaborating personnel or mobile devices obtained in step 110, the edge controller uses dead reckoning or kinematic prediction models to deduce their predicted spatial coordinates at each moment within a preset time window. Based on the predicted coordinates at each moment, a corresponding three-dimensional occupied region is constructed (e.g., simplifying the collaborating personnel into a dynamic capsule envelope that moves over time and has a safety margin), forming a dynamic spatial constraint boundary that changes dynamically over time. .

[0059] Spacetime boundary fusion: The static spatial constraint boundary and the dynamic spatial constraint boundary are superimposed in union on the time axis of a preset time window to generate a unified spatiotemporal constraint boundary. :

[0060] Furthermore, in the virtual workspace, the edge controller overlays static environmental elements and dynamic obstacle information on the timeline to construct spatiotemporal constraint boundaries. This process can be the initial construction during system initialization, or it can be an incremental update based on the previously constructed historical spatiotemporal constraint boundaries and the newly perceived physical environmental elements during robot operation.

[0061] Specifically, during the initial construction, the edge controller retrieves the complete static obstacle geometric coordinates and the current dynamic obstacle trajectory, and generates the initial spatiotemporal constraint boundary according to the following steps. During operation, when the external 3D sensing device detects the appearance of new static obstacles or changes in the trajectory of dynamic obstacles in the workspace, the edge controller performs local correction or replacement of the corresponding area in the existing spatiotemporal constraint boundary without full reconstruction, thereby improving the real-time performance of boundary updates.

[0062] After obtaining the spatiotemporal constraint boundary, the edge controller maps the predicted running trajectory output in step 130 to the virtual workspace for collision interference detection, generating a collision risk evolution curve. In a specific embodiment, this step is implemented by the following steps: Calculate the shortest distance between the robot body and the spatiotemporal constraint boundary at each moment when the robot moves along the predicted running trajectory within a preset time window; The shortest distance is compared with the preset safe distance, which is then converted into a collision risk value distributed over time, generating a collision risk evolution curve.

[0063] Specifically, because robots are physical entities, the edge controller simplifies the physical outlines of the robot's joints and links into a set of interconnected geometric envelopes (such as capsules or cylinders). This is applied to each projection moment within a pre-defined future time window. This drives the simplified robot body along the predicted trajectory. Motion. Calculation at time [time]. The geometric envelope of the robot body and the spatiotemporal constraint boundary at the current moment. The shortest three-dimensional Euclidean distance between The edge controller is preset with a safe distance. (e.g., 200 mm) and a critical collision distance (For example, 50 mm). Calculate the shortest distance at each time step. Transformed into collision risk values ​​distributed over time. The conversion rules are as follows:

[0064] when ≥ hour, = 0 indicates safety; when < < hour, This indicates a risk of collision. when ≤ hour, =1 indicates an extremely high risk or that a collision has already occurred.

[0065] Risk values ​​calculated for all simulation moments Arranged along a future timeline, this generates a continuously distributed collision risk evolution curve. The curve clearly predicts at what point in a future task (e.g., 10 minutes) the robot will be too close to a person or obstacle, and what percentage of the probability of a collision will occur.

[0066] Step 150: Perform a comprehensive state assessment based on the trajectory deviation curve and the collision risk evolution curve. When the operational risk is determined to exceed the preset safety threshold, calculate the compensation control quantity and the collision avoidance bias quantity and inject them into the motion control loop to achieve proactive safety intervention for the robot.

[0067] This step comprehensively assesses the robot's operational risks based on the trajectory deviation curve and collision risk evolution curve calculated in steps 130 and 140, and injects corrections into the motion control loop before future deviations or collisions actually occur, thus achieving proactive intervention.

[0068] In a specific implementation, step 150 is achieved through the following steps: Specifically, comprehensive assessment and control decisions are achieved collaboratively through the following two branches.

[0069] In the first scenario, the trajectory deviates beyond the limit but there is no risk of collision.

[0070] If the trajectory deviation curve exceeds the trajectory deviation threshold but the collision risk evolution curve does not exceed the collision risk threshold, then the operational risk is determined to exceed the preset safety threshold. The compensation control quantity is calculated based on the deviation between the predicted running trajectory and the expected motion command, and then injected into the motion control loop to correct the running trajectory. When the edge controller determines within a preset time window that the trajectory deviates from the curve Exceeding the preset trajectory deviation threshold And the collision risk evolution curve Not exceeding the safety collision risk threshold If the current operational risk falls under the "accuracy degradation but space safety" state, then active compensation for the out-of-tolerance trajectory is executed. The edge controller calculates the compensation torque required to overcome joint elastic deformation and friction increments based on the trajectory deviation at that moment and the current stiffness degradation. The compensation torque As a feedforward correction torque, it is directly superimposed on the torque control loop of the underlying servo driver of the physical robot. By increasing the joint active torque, it compensates for the trajectory deviation caused by mechanical decay and corrects the running trajectory.

[0071] The second scenario is that the collision risk exceeds the limit.

[0072] If the collision risk evolution curve exceeds the collision risk threshold, the operational risk is determined to exceed the preset safety threshold. The collision avoidance bias is calculated based on the shortest distance direction between the predicted trajectory and the spatiotemporal constraint boundary, and then injected into the motion control loop to achieve emergency collision avoidance.

[0073] When the edge controller determines the collision risk evolution curve Exceeding the safety collision risk threshold This indicates that the robot will face a collision risk with a person or obstacle at some point in the future. In this situation, regardless of trajectory accuracy, a safe avoidance maneuver must be executed. The current operational risk is classified as a "space collision crisis" state, triggering emergency collision avoidance. The edge controller extracts the shortest distance vector between the robot body and the spatiotemporal constraint boundary at that moment, which is the unit direction normal vector from the nearest point on the boundary to the nearest point on the robot body. Based on the magnitude of the shortest distance, the collision avoidance angle offset used to keep the robot body away from the spatiotemporal constraint boundary is calculated. :

[0074] in, For collision avoidance gain coefficient, To establish a safe distance, This represents the shortest distance at the current moment. This formula causes the robot to generate a "virtual repulsion displacement" away from the obstacle.

[0075] The calculated collision avoidance bias The trajectory planning layer of the robot motion controller is injected in real time to perform local smoothing corrections on the original command trajectory, enabling the robot to bypass the spatiotemporal constraint boundaries and achieve advanced emergency collision avoidance.

[0076] Based on the same inventive concept, as an implementation of the above-mentioned robot state prediction method based on digital twins, this application also provides a robot state prediction system based on digital twins. Figure 2 This is a structural diagram of the system in an embodiment of this application. See also... Figure 2 As shown, the system includes: The acquisition module is used to acquire the robot's real-time operating data and the physical environmental elements of its workspace; The decay parameter update module is used to evaluate the time-varying mechanical decay of the robot based on real-time operating data and update the decay parameters in the digital twin to obtain a digital twin that is synchronized in real time with the current electromechanical loss state. The digital twin is a digital model pre-built in the edge controller that corresponds to the robot's mechanical dynamics properties. The calculation module is used to obtain the robot's expected motion commands within a preset time window, input the decay prediction trend and expected motion commands within the preset time window into the digital twin for advanced extrapolation, and calculate the predicted running trajectory and trajectory deviation curve; wherein, the decay prediction trend is extrapolated by the digital twin based on historical decay parameters; The collision risk detection module is used to construct spatiotemporal constraint boundaries in the virtual workspace based on physical environment elements, map the predicted running trajectory to the virtual workspace to perform collision risk detection, and generate collision risk evolution curves. The risk control module is used to perform a comprehensive state assessment based on the trajectory deviation curve and the collision risk evolution curve. When the operational risk is determined to exceed the preset safety threshold, the module calculates the compensation control quantity and the collision avoidance bias quantity and injects them into the motion control loop to achieve safe intervention for the robot.

[0077] This application effectively solves the problem of inaccurate prediction and disconnected safety intervention in existing technologies by introducing a unified decision-making mechanism for decay prediction trends and internal and external risks. On the one hand, the edge controller uses the accumulated decay parameter updates within historical operating cycles to extrapolate the continuous change curves of stiffness attenuation factor, viscous friction correction coefficient, and Coulomb friction correction coefficient within a preset time window through a time series prediction algorithm. These curves are then injected as time-varying dynamic parameters into the dynamic equations of the digital twin, enabling the predicted operating trajectory to reflect the continuous evolution of mechanical decay on the prediction time axis. This avoids the accumulation of prediction deviations caused by parameter freezing and significantly improves the accuracy of long-cycle trajectory prediction. On the other hand, by incorporating the trajectory deviation curve and collision risk evolution curve into a unified comprehensive state assessment framework, and setting trajectory deviation thresholds and collision risk thresholds, the application classifies and determines two risk states: "accuracy decay" and "collision crisis." When the trajectory deviation exceeds the limit but the collision risk does not, compensation control is triggered; when the collision risk exceeds the limit, emergency collision avoidance is triggered first. This hierarchical and mutually exclusive decision-making mechanism avoids the problems of conflicting control commands or delayed intervention caused by decision-making based on internal and external risks, thus ensuring the robot's operational accuracy and inherent safety in complex human-machine collaborative environments.

[0078] This application also provides an electronic device, comprising: a processor, a memory, and a computer program stored in the memory and executable on the processor, such as a program for a robot state prediction method based on digital twins. When the processor executes the computer program, it implements all the steps in the aforementioned robot state prediction method based on digital twins, such as S110 to S150 and their sub-steps. Alternatively, when the processor executes the computer program, it implements the functions of each module / unit in the aforementioned system embodiments.

[0079] For example, the computer program may be divided into one or more modules / units, which are stored in the memory and executed by the processor to complete this application. The one or more modules / units may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program in the electronic device.

[0080] The electronic device may be a computing device such as a server or cloud server deployed in a robot state prediction system based on digital twins. The electronic device may include, but is not limited to, processors and memory. Those skilled in the art will understand that the above components are merely examples of electronic devices and do not constitute a limitation on the electronic device. It may include more or fewer components than described above, or combine certain components, or different components. For example, the electronic device may also include data acquisition cards and output cards for connecting sensors and actuators, network communication devices, buses, etc.

[0081] The processor can be a central processing unit, or other general-purpose processors, digital signal processors, application-specific integrated circuits (ASICs), off-the-shelf programmable gate arrays (OPGs) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the electronic device, connecting various parts of the entire electronic device via various interfaces and lines.

[0082] The memory can be used to store the computer programs and / or modules. The processor implements various functions of the electronic device by running or executing the computer programs and / or modules stored in the memory and by calling data stored in the memory. The memory may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function, etc.; the data storage area may store data created according to system usage, etc. In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, RAM, plug-in hard disk, smart memory card, secure digital card, flash memory card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0083] Wherein, if the modules / units integrated in the electronic device are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, a read-only memory, a random access memory, an electrical carrier signal, a telecommunication signal, and a software distribution medium, etc.

[0084] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A robot state prediction method based on digital twins, characterized in that, The method includes: Acquire real-time operational data of the robot and the physical environmental elements of its workspace; The time-varying mechanical decay of the robot is evaluated based on the real-time operating data, and the decay parameters in the digital twin are updated to obtain a digital twin that is synchronized in real time with the current electromechanical loss state. The digital twin is a digital model pre-built in the edge controller that corresponds to the mechanical dynamics properties of the robot. The robot's expected motion commands within a preset time window are obtained. The decay prediction trend within the preset time window and the expected motion commands are input into the digital twin for advanced extrapolation, and the predicted running trajectory and trajectory deviation curve are calculated. The decay prediction trend is obtained by extrapolation from the digital twin based on historical decay parameters. Based on the physical environment elements, a spatiotemporal constraint boundary is constructed in the virtual workspace, and the predicted running trajectory is mapped to the virtual workspace for collision risk detection, generating a collision risk evolution curve. A comprehensive state assessment is performed based on the trajectory deviation curve and the collision risk evolution curve. When the operational risk is determined to exceed the preset safety threshold, the compensation control quantity and collision avoidance bias quantity are calculated and injected into the motion control loop to achieve safe intervention for the robot.

2. The method according to claim 1, characterized in that, The acquisition of the robot's real-time operating data and the physical environmental elements of its workspace includes: The real-time operating data includes joint angular displacement, joint angular velocity, joint contact temperature, and q-axis current of the servo motor, all collected by the robot's joint sensors and servo drivers. The physical environment elements include the spatial coordinates of static obstacles and the motion trajectories of dynamic obstacles, which are collected in real time by external three-dimensional sensing devices deployed in the workspace.

3. The method according to claim 2, characterized in that, The process of evaluating the time-varying mechanical decay of the robot based on the real-time operational data and updating the decay parameters within the digital twin includes: The theoretical torque is calculated based on the joint angular displacement and joint angular velocity, and the actual torque is calculated based on the q-axis current of the servo motor, thus obtaining the torque deviation between the two. By combining the joint contact temperature, the interference caused by temperature fluctuations in the torque deviation is eliminated, and the stiffness degradation and viscous friction increment excluding the influence of temperature are obtained. Extract the high-frequency fluctuation characteristics in the q-axis current of the servo motor to determine the Coulomb friction increment caused by gear wear; Based on the stiffness degradation, viscous friction increment, and Coulomb friction increment, the corresponding model parameter update amounts are calculated and updated in the digital twin.

4. The method according to claim 1, characterized in that, The recession forecast trend is obtained through the following methods: Obtain the cumulative amount of decay parameter updates accumulated by the digital twin during its historical operating cycle; The historical sequence of the decay parameter update is fitted using a time series prediction algorithm, and the parameter change curve that evolves over time within the preset time window is extrapolated as the decay prediction trend.

5. The method according to claim 1, characterized in that, The step of inputting the predicted decline trend within the preset time window and the expected movement command into the digital twin for advanced extrapolation, and calculating the predicted trajectory and trajectory deviation curve, includes: The predicted decay trend is substituted into the dynamic equation of the digital twin as a time-varying decay parameter. The expected motion command is used as input, and simulation is performed within the preset time window to obtain the predicted joint motion sequence, which is used as the predicted running trajectory. The deviation between the predicted trajectory and the expected motion command is calculated, and the trajectory deviation curve is generated as a function of time.

6. The method according to claim 2, characterized in that, The construction of spatiotemporal constraint boundaries in the virtual workspace based on the physical environment elements includes: The spatial coordinates of the static obstacle are converted into static spatial constraint boundaries in the virtual workspace; Based on the movement trajectory of the dynamic obstacle, predict the dynamic occupied area within a future preset time window, and construct a dynamic spatial constraint boundary that changes dynamically over time. The static spatial constraint boundary and the dynamic spatial constraint boundary are merged on the time axis to generate the spatiotemporal constraint boundary.

7. The method according to claim 1, characterized in that, The step of mapping the predicted trajectory to the virtual workspace for collision risk detection and generating a collision risk evolution curve includes: Calculate the shortest distance between the robot body and the spatiotemporal constraint boundary at each moment when the robot moves along the predicted running trajectory within the preset time window; The shortest distance is compared with the preset safe distance and converted into a collision risk value distributed over time, thus generating the collision risk evolution curve.

8. The method according to claim 1, characterized in that, The step of performing a comprehensive state assessment based on the trajectory deviation curve and the collision risk evolution curve, and calculating the compensation control quantity and collision avoidance bias quantity and injecting them into the motion control loop when the operational risk is determined to exceed a preset safety threshold, includes: If the trajectory deviation curve exceeds the trajectory deviation threshold and the collision risk evolution curve does not exceed the collision risk threshold, then the operation risk is determined to exceed the preset safety threshold. The compensation control quantity is calculated based on the deviation between the predicted running trajectory and the expected motion command, and injected into the motion control loop to correct the running trajectory; or, If the collision risk evolution curve exceeds the collision risk threshold, the operational risk is determined to exceed the preset safety threshold. The collision avoidance bias is calculated based on the shortest distance direction between the predicted trajectory and the spatiotemporal constraint boundary, and injected into the motion control loop to achieve emergency collision avoidance.

9. A robot state prediction system based on digital twins, deployed in the edge controller of a robot, characterized in that, The system includes: The acquisition module is used to acquire the robot's real-time operating data and the physical environmental elements of its workspace; The decay parameter update module is used to evaluate the time-varying mechanical decay degree of the robot based on the real-time operating data and update the decay parameters in the digital twin to obtain a digital twin that is synchronized with the current electromechanical loss state in real time. The digital twin is a digital model pre-built in the edge controller that corresponds to the robot's mechanical dynamics attributes. The calculation module is used to obtain the robot's expected motion command within a preset time window, input the decay prediction trend within the preset time window and the expected motion command into the digital twin for advanced extrapolation, and calculate the predicted running trajectory and trajectory deviation curve; wherein, the decay prediction trend is obtained by extrapolation from the digital twin based on historical decay parameters; The collision risk detection module is used to construct a spatiotemporal constraint boundary in the virtual workspace based on the physical environment elements, map the predicted running trajectory to the virtual workspace to perform collision risk detection, and generate a collision risk evolution curve. The risk control module is used to perform a comprehensive state assessment based on the trajectory deviation curve and the collision risk evolution curve. When it is determined that the operating risk exceeds the preset safety threshold, the module calculates the compensation control quantity and the collision avoidance bias quantity and injects them into the motion control loop to achieve safe intervention for the robot.

Images